One of the challenges faced in image classification is that the appearance of objects is dynamic. Just as there are many different breeds of cats and dogs, there are an infinite number of ways cats and dogs can appear in images. This makes it difficult for rudimentary image classification techniques, as it is impossible to show an infinite number of photos of cats and dogs to a computer.
However, this really shouldn't be a problem at all. Humans don't require an infinite number of photos of cats and dogs to differentiate between the two. A toddler can easily differentiate cats and dogs once he has seen just a few of them. If we think about how humans approach image classification, we notice that humans tend to look for landmark features while trying to identify an object. For example, we know that cats tend to be smaller in size compared to dogs...
